Advancements in Histopathological Image Analysis for Accurate Breast Cancer Classification: Leveraging Self-Supervised Contrastive Learning and Transfer Learning

• DOI: 10.2139/ssrn.4530186
• Publication Date: 2023-08-09T18:37:21Z
• Authors: Faisal Bin Ashraf; S. M. Maksudul Alam; Shahriar M. Sakib
• Keywords: Transfer of learning

Breast cancer is a prevalent and potentially life-threatening disease among women worldwide. The interpretation of histopathological images by pathologists, although crucial for early detection, labor-intensive task prone to errors. Consequently, there pressing need reliable automated methods aid in breast detection classification. In this research, our objective was effectively identify classify malignant benign using images. We pursued two approaches address challenge. Firstly, we employed self-supervised contrastive learning enhance the process. Secondly, leveraged transfer combining blocks from ResNet50 naive inception model improve architecture. Through meticulous application data augmentation techniques block-by-block fine-tuning on with varying magnification levels, achieved significant performance improvements. Our proposed deep architecture, incorporating fine-tuning, surpassed state-of-the-art accuracy classifying cancerous Notably, trained architecture rates 99%, 96%, 98%, 96% 40X, 100X, 200X, 400X magnified images, respectively. Overall, research presents promising approach classification, demonstrating potential improved image analysis.